sciphys-ai Analyze Instrument Data
Build a traceable analysis from raw measurements. Treat the instrument-specific implementation as a plugin; keep the evidence workflow stable across modalities.
Workflow
- Inventory every input and preserve the original files unchanged.
- Identify modality, vendor format, acquisition metadata, units, calibration, and sample context. State unknowns instead of guessing.
- Hash inputs with
scripts/build_evidence_manifest.pyand start an Evidence Bundle. - Select a compatible parser/analyzer from
registry/plugins/. If none exists, produce a gap report rather than fabricating results. - Run quality checks before interpretation: parse completeness, unit consistency, saturation, missing ranges, duplicates, baseline/noise, and modality-specific artifacts.
- Produce machine-readable results plus a short scientific interpretation. Separate observation, derived result, interpretation, and hypothesis.
- Link every claim to result or reference nodes and record parameters, software versions, warnings, and alternative explanations.
- Validate the bundle with
sciphys evidence validate <bundle.json>.
Outputs
Return or create:
- an input manifest with SHA-256 hashes;
- normalized data or an explicit unsupported-format report;
- quality-control findings and analysis parameters;
- structured results with units and uncertainty where available;
- an Evidence Bundle connecting inputs, processes, outputs, and claims;
- concise next steps that distinguish required checks from optional follow-up.
Read references/evidence-policy.md before assigning confidence or making cross-instrument claims.
Guardrails
- Never infer an instrument setting that materially changes the result.
- Never overwrite raw data.
- Never call a synthetic regression fixture a scientific benchmark.
- Flag reference-database licensing and provenance.
- Require expert review for safety-critical, clinical, regulatory, or publication-final conclusions.